Zipeng Zhu

I am a quantitative data scientist with strong foundations in statistical modeling, supervised learning, and probability theory. I focus on developing rigorously validated, interpretable models that support real-world decision making.

My experience includes offline recommendation modeling where I framed engagement as a supervised ranking problem, built leakage-aware temporal cross-validation pipelines over 500,000+ interactions, and benchmarked graph-based and matrix-factorization methods against statistical baselines.

I have built time-series forecasting systems combining ARIMA, gradient-boosted trees, and deep learning (LSTM), applying rolling-origin validation and bias–variance analysis to improve forecast stability and quantify uncertainty for decision support.

I also design user risk and anomaly detection pipelines using engineered behavioral features, tree-based classifiers, and Isolation Forests, emphasizing precision–recall calibration, interpretable risk scores, and operational robustness through sensitivity analysis.